Enhanced significant wave height prediction in the Southern Ocean using an ANFIS model optimized with subtractive clustering

Accurate prediction of significant wave height (SWH) in the Southern Ocean remains a critical challenge due to extreme weather conditions and limited observational data, impacting maritime safety and climate research. This study introduces an Adaptive Neuro-Fuzzy Inference System (ANFIS) optimized with subtractive clustering for SWH forecasting, with its novelty lying in the integration of sequential time-lagged inputs and automated fuzzy rule generation to model complex nonlinear wave dynamics. The model leverages marine meteorological variables, including mean sea level pressure (MSLP), surface wind speed (SWS), and historical SWH records, using three consecutive time steps. High-quality data from MetOcean Spotter buoys and ERA5 reanalysis, collected at 3-hour intervals during 2019-2020, were divided into training (70%), testing (30%), and an independent prediction set (250 samples). Compared to conventional regression techniques such as neural networks, support vector machines, and Gaussian process regression, the ANFIS model demonstrated superior performance, achieving a root mean square error (RMSE) of 0.5142 m and a coefficient of determination (R²) of 0.8948. With efficient computational performance (65 milliseconds per prediction) and interpretable fuzzy rules, this model offers a robust tool for operational wave forecasting in polar regions, advancing maritime safety and enhancing understanding of ocean-atmosphere interactions.